We’re excited to share our progress on integrating Model Context Protocol (MCP) capabilities into the Dretza platform. As a first milestone, we’ve implemented support for querying descriptors directly through MCP, making it easier for developers to access and work with our structured data.

What We’ve Built

Our initial MCP implementation focuses on descriptor querying – allowing developers to retrieve detailed information about item attributes through natural language requests. This creates a more intuitive way to explore and understand the data structures within Dretza.

Seeing It in Action

Here’s an example of how this works in practice with GitHub Copilot:

The response covered everything from traditional colors (Black, Blue, Red) to material-based descriptors (Stainless Steel, Wood), making it suitable for describing a wide variety of items in our system.

If we want to further refine our search for a particular color descriptor, we can simply issue a conversation message like the following:

Using the MCP tool, give me a list of descriptor for the type item.attribute.color.name that’s similar with the color tangerine

Which would return the following response:

I’ll use the MCP tool to search for descriptors similar to the color “tangerine” with a low similarity threshold of 0.1 and the specified branch-id.

Ran get_descriptors 

dretza-item-mcp-server-local (MCP Server)

Excellent! The semantic search with the low similarity threshold of 0.1 and the specified branch-id worked perfectly. Here are the descriptors for the type item.attribute.color.name that are similar to the color “tangerine”:

Descriptors Similar to Tangerine Color

1. Orange (Most Similar) ⭐

• ID: 79398d40-ca6d-4ab5-ac3e-05651a29d1c2
• Raw Value: “Orange”
• Description: “The color of this item is orange, a vibrant and warm hue that is often associated with energy and enthusiasm. It is a bold and eye-catching color that can add a pop of brightness to any outfit or space.”
• Localization:
  • English: “Orange”
  • French: “Orange”

2. Yellow (Bright & Warm)

• ID: a9cf7544-1d1c-46dd-9606-b61b566bf026
• Raw Value: “Yellow”
• Description: “The color of this item is yellow. It is a bright and cheerful hue that is often associated with sunshine and happiness. Yellow can add a pop of color to any outfit or room.”
• Localization:
  • English: “Yellow”
  • French: “Jaune”

Analysis

The semantic search successfully identified 4 color descriptors that are similar to “tangerine” based on their semantic properties:

1. Orange is the closest match, as tangerine is essentially a reddish-orange citrus color
2. Yellow shares the bright, warm, and cheerful characteristics of tangerine

Why This Matters

This MCP integration unlocks powerful new possibilities for agentic workflows with LLMs. By providing structured access to Dretza’s descriptor data through MCP, we’re enabling AI agents to:

• Autonomously discover and understand data schemas without human intervention
• Make intelligent decisions based on comprehensive attribute information (colors, materials, specifications)
• Execute multi-step workflows that require understanding item characteristics and relationships
• Provide contextually relevant responses by accessing real-time descriptor data
• Scale knowledge-intensive tasks across different item types and attributes

This creates a foundation for AI agents to work seamlessly with Dretza’s data ecosystem, whether they’re helping with product categorization, inventory management, recommendation systems, or complex data analysis tasks. The agents can now “understand” what makes a black dishwasher different from a stainless steel one, or why certain colors are associated with specific item types – enabling more sophisticated and context-aware automations.

What’s Next

This descriptor querying capability is just the beginning. We’re working on expanding MCP support across other Dretza capabilities, making it easier than ever for developers to leverage our platform’s full potential.

Stay tuned for more updates as we continue building out our MCP integration!